Hamburger moment problem for powers and products of random variables

Jordan Stoyanov, Gwo Dong Lin, Anirban DasguptaJournal of Statistical Planning and Inference 2013 · DOI

TL;DR

This paper delves into the "moment determinacy" of random variables, a property that dictates whether a distribution is uniquely defined by its moments. The authors develop new theoretical tools to analyze this property for powers and products of random variables, providing a comprehensive framework to determine when these transformations lead to moment-indeterminate distributions, a common occurrence in real-world modeling.

What Problem Does This Paper Address?

The core problem this paper tackles is understanding the "moment determinacy" of probability distributions, particularly when those distributions are transformed through powers or products of independent random variables. In essence, if we know all the moments (mean, variance, skewness, etc.) of a random variable, does that uniquely identify its underlying probability distribution? This is known as the Hamburger moment problem. While this problem has a rich history, analyzing the moment determinacy of transformed random variables (like powers or products) is significantly more complex. These transformations are crucial because many real-world phenomena involve such operations (e.g., signal processing, financial modeling). The paper aims to provide general theorems and specific examples to characterize when these transformations result in distributions that are not uniquely determined by their moments (moment-indeterminate), which is a common and often problematic scenario in statistical modeling.

Methods

The paper introduces and utilizes several key mathematical concepts and techniques:

  • Growth Rate of Moments: A central theme is analyzing the "growth rate" of the even-order moments of a random variable. This is quantified by the ratio of consecutive even moments, $m_{2(k+1)}/m_{2k}$.
  • Carleman's and Cramér's Conditions: These are classical conditions used to determine moment determinacy. The paper establishes how the growth rate of moments relates to these conditions.
  • Lyapunov's Inequality and Hölder's Inequality: These fundamental inequalities are used to establish relationships between moments.
  • Krein's Condition and Logarithmic Integral: The paper leverages these conditions, particularly for analyzing the tails of distributions, which are crucial for moment indeterminacy.
  • Stirling's Approximation: This approximation for the Gamma function is used extensively to analyze the asymptotic behavior of moments.
  • General Theorems: The paper presents several general theorems (Theorems 1-3) that link the growth rate of moments to moment determinacy (both for determinacy and indeterminacy).
  • Specific Distribution Analysis: The authors apply these general theorems to analyze the moment determinacy of powers and products of specific distributions like the Double Generalized Gamma (DGG), Normal, Laplace, and Logistic distributions.
  • Inductive Proofs: For products of multiple random variables, inductive arguments are used to extend results from two variables to n variables.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Growth Rate is Key: The "growth rate" of the even-order moments is a powerful indicator of moment determinacy. Slow growth (e.g., polynomial) often implies determinacy, while fast growth (e.g., exponential or faster) suggests indeterminacy.
  • Powers vs. Products: The paper explores how the moment determinacy of a random variable relates to the determinacy of its powers and products. Interestingly, these properties are not always shared, especially for distributions like the Laplace.
  • "Heavy Tails" Lead to Indeterminacy: The paper formalizes the intuitive idea that distributions with "heavy tails" (where extreme values are more probable than in lighter-tailed distributions) are more likely to lead to moment-indeterminate transformations.
  • New Proofs for Classic Results: The authors provide new, more transparent proofs for some known results, such as the moment indeterminacy of products of three or more independent normal random variables.
  • Random Exponents and Products: The paper extends the analysis to "random powers" and "random products" where the number of factors or the exponent is itself a random variable, showing that these can also lead to moment indeterminacy.
  • Unconventional Combinations: The study includes products of independent random variables of different types (e.g., Laplace and Logistic), revealing that such combinations can also be moment-indeterminate.

Key Results

  • Theorems 1 & 2: Establish that if the ratio of even moments grows at a certain polynomial rate ($m_{2(k+1)}/m_{2k} = O((k+1)^2)$), the variable is moment-determinate (satisfies Cramér's condition).
  • Theorem 3: Shows that if the ratio of even moments grows faster than a certain polynomial rate ($m_{2(k+1)}/m_{2k} \ge c(k+1)^{2+\epsilon}$), and the density has specific tail behavior, the variable is moment-indeterminate.
  • Corollaries & Propositions: Provide specific results for powers and products of i.i.d. random variables, showing conditions under which they are moment-determinate or indeterminate. For instance, products of n independent normal random variables are moment-indeterminate for $n \ge 3$.
  • Logistic Distribution: Powers of a standard logistic variable are moment-determinate for orders 1 and 2, but indeterminate for odd orders $\ge 3$. Products of two or more independent logistic variables are always moment-indeterminate.
  • Double Generalized Gamma (DGG) Distribution: The paper provides detailed analysis for this flexible family, showing how parameters influence the moment determinacy of powers and products.

Limitations and Future Work

  • Focus on Symmetric Distributions: Many of the results, particularly for products, rely on the assumption of symmetric densities. Extending these to asymmetric distributions would be a natural next step.
  • Specific Tail Conditions: Theorem 3 requires specific conditions on the tail behavior of the density for moment indeterminacy. Exploring the precise boundaries of these conditions for various distributions could be fruitful.
  • More Complex Transformations: The paper focuses on powers and products. Investigating other nonlinear transformations (e.g., sums of products, or more complex functions of random variables) could be a direction for future research.
  • Numerical Verification: While theoretical results are provided, numerical simulations could further illustrate and verify these findings for specific distributions and parameters.

Summary updated

BibTeX
@article{stoyanov2013hamburger,
  author = {Jordan Stoyanov and Gwo Dong Lin and Anirban Dasgupta},
  title = {{Hamburger moment problem for powers and products of random variables}},
  journal = {Journal of Statistical Planning and Inference},
  year = {2013},
  doi = {10.1016/j.jspi.2013.11.002},
}
Figure